What Coherent ZR Is Eating in Long-Haul (And What It Still Can't Touch)

Pluggable ZR did to metro DCI what nobody expected at this pace. Between roughly 2022 and 2025, transponder shelves got pulled out of metro builds across most of the hyperscalers, replaced by 400ZR plugged directly into the router. The economics weren't even close. A QSFP-DD ZR plus a 1RU router slot beats a transponder shelf on capex, opex, footprint, and lead time. So the transponder shelf moved.

It did not die. It got niche. The interesting question now is what stays, what is still under threat, and where the next plug-versus-shelf fight is.

Where pluggable ZR ate the shelf, and stayed eating

Metro DCI under 120 km, with line-side OSNR comfortably above margin, was a slam-dunk. 400ZR in a host router replaced the old transponder-plus-grey-transceiver pair, and the savings paid for themselves in a quarter. Some hyperscalers reported 60 to 70 percent capex savings on their metro builds. The order book confirms it — public filings from two major optical houses show their metro transponder revenue down significantly year on year since 2023.

Regional DCI, in the 200 to 400 km range, was the next domino. 400ZR+ with its higher launch power and amplification options reached most of that distance with practical margin. Add a single inline amplifier on a clean fibre and you are at 400 km without a transponder anywhere. Order data through 2025 shows that range is mostly migrated.

What's still transponder territory

Submarine. The reach math doesn't bend. You need a transponder-grade DSP, optimised amplifier chains, and the kind of nonlinearity compensation that lives in transponder cards because it costs too much power to put in a pluggable. 400ZR+ at 800 km on a clean terrestrial fibre? Yes, with margin. 800 km of submarine cable with cascaded EDFAs? Not yet.

Ultra-long-haul terrestrial — 1,500-plus km, multiple amp huts, mixed fibre types. Same answer. The DSP in a pluggable has to work inside a 5 W cage, which constrains how aggressive the nonlinearity comp and the constellation shaping can be. The transponder does not have that constraint, and the reach advantage shows up in the link budget.

Transoceanic. Different conversation entirely. The shelf isn't even competing with pluggables there — it's competing with custom-developed line systems and proprietary subsea DSPs. The pluggable doesn't enter this market for at least another generation.

ROADM passthrough. If your link traverses three or more wavelength-selective switches, the per-pass insertion loss and filtering effects punish the lower-OSNR pluggable. You can engineer around it (lower channel count, custom shaping), but the operational complexity is what kills it. Most hyperscalers and most carriers keep transponders on those routes.

The fight that's actually happening right now

The interesting front line in 2026 is not ZR versus shelf. ZR won the easy ones. The interesting front line is 800ZR+ versus the embedded transponder DSP for the 400 to 800 km range.

800ZR+ samples are out. Pre-production volume is in a handful of hyperscaler trials. The reach math, on paper, takes a real bite out of regional and the lower end of long-haul. The honest answer on whether it makes it into production depends on three things:

• Whether the DSP power can be held under 6 W in a QSFP-DD cage at the launch power 800ZR+ requires. This is the binding constraint.
• Whether OIF specifications for 800ZR+ converge fast enough that multi-vendor interop is real before the embedded transponders ship their next generation.
• Whether the hyperscaler order book moves before the transponder vendors counter-price.

My read, based on lead times for the silicon and public messaging from the three remaining DSP houses, is that 800ZR+ takes meaningful share in 2027, not 2026. Vendors will say otherwise.

What this means if you're planning capacity now

If you are building DCI for the next two years and your link plan has anything between metro and regional reach, 400ZR or 400ZR+ is the default. Don't let a transponder salesperson convince you otherwise. The price-performance is not close.

If you are planning routes that need 800-plus km of clean fibre, multiple ROADM passes, or any submarine segment, the transponder shelf still wins, and the math will be unchanged through 2027.

If your routes are in the 400 to 800 km regional band and you're refreshing in late 2026 or 2027 — that's the band where 800ZR+ will land or won't land, and your forecast assumption matters. Default to transponder for one more cycle if you need it ordered tomorrow. Default to 800ZR+ if you have eighteen months of design runway and don't mind being on the leading edge of multi-vendor interop.

The shelf is not dead. It got specific.